From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754788AbcBWRyc (ORCPT ); Tue, 23 Feb 2016 12:54:32 -0500 Received: from mx1.redhat.com ([209.132.183.28]:41374 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754292AbcBWRyN (ORCPT ); Tue, 23 Feb 2016 12:54:13 -0500 From: Prarit Bhargava To: linux-kernel@vger.kernel.org Cc: Prarit Bhargava , "Cc: Petr Mladek" , John Stultz , Xunlei Pang , Thomas Gleixner , Baolin Wang , Andrew Morton , Greg Kroah-Hartman , Petr Mladek , Tejun Heo , Peter Hurley , Vasily Averin , Joe Perches Subject: [PATCH 2/2 v6] printk, allow different timestamps for printk.time Date: Tue, 23 Feb 2016 12:54:00 -0500 Message-Id: <1456250040-22351-3-git-send-email-prarit@redhat.com> In-Reply-To: <1456250040-22351-1-git-send-email-prarit@redhat.com> References: <1456250040-22351-1-git-send-email-prarit@redhat.com> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Over the past years I've seen many reports of bugs that include time-stamped kernel logs (enabled when CONFIG_PRINTK_TIME=y or print.time=1 is specified as a kernel parameter) that do not align with either external time stamped logs or /var/log/messages. This also makes determining the time of a failure difficult in cases where /var/log/messages is unavailable. For example, [root@intel-wildcatpass-06 ~]# date; echo "Hello!" > /dev/kmsg ; date Thu Dec 17 13:58:31 EST 2015 Thu Dec 17 13:58:31 EST 2015 which displays [83973.768912] Hello! on the serial console. Running a script to convert this to the stamped time, [root@intel-wildcatpass-06 ~]# ./human.sh | tail -1 [Thu Dec 17 13:59:57 2015] Hello! which is already off by 1 minute and 26 seconds off after ~24 hours of uptime. This occurs because the time stamp is obtained from a call to local_clock() which (on x86) is a direct call to the hardware. These hardware clock reads are not modified by the standard ntp or ptp protocol, while the other timestamps are, and that results in situations external time sources are further and further offset from the kernel log timestamps. This patch introduces printk.time=[0-3] allowing a user to specify an adjusted clock to use with printk timestamps. The hardware clock, or the existing functionality, is preserved by default. Real clock & 32-bit systems: Selecting the real clock printk timestamp may lead to unlikely situations where a timestamp is wrong because the real time offset is read without the protection of a sequence lock in the call to ktime_get_log_ts() in printk_get_ts(). [v2]: use NMI safe timekeeping access functions [v3]: use tglx's ktime_get_log_ts() suggestion to get monotonic & real clocks [v4]: Fix kernel parameters [v5]: Modify configs to accept [0-3] [v6]: Implement Kconfig changes from Petr Mladek, fix CONFIG_PRINTK_TIME=0 and dmesg bug Cc: Cc: Petr Mladek Cc: John Stultz Cc: Xunlei Pang Cc: Thomas Gleixner Cc: Baolin Wang Cc: Andrew Morton Cc: Greg Kroah-Hartman Cc: Petr Mladek Cc: Tejun Heo Cc: Peter Hurley Cc: Vasily Averin Cc: Joe Perches Signed-off-by: Prarit Bhargava --- Documentation/kernel-parameters.txt | 6 ++++-- include/linux/timekeeping.h | 1 + kernel/printk/printk.c | 35 +++++++++++++++++++++++++++++++++-- kernel/time/timekeeping.c | 13 +++++++++++++ lib/Kconfig.debug | 7 ++++--- 5 files changed, 55 insertions(+), 7 deletions(-) diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt index a86642c..92e9076 100644 --- a/Documentation/kernel-parameters.txt +++ b/Documentation/kernel-parameters.txt @@ -3065,8 +3065,10 @@ bytes respectively. Such letter suffixes can also be entirely omitted. Format: (1/Y/y=enable, 0/N/n=disable) default: disabled - printk.time= Show timing data prefixed to each printk message line - Format: (1/Y/y=enable, 0/N/n=disable) + printk.time= Show timestamp prefixed to each printk message line + Format: + (0/N/n = disable, 1/Y/y = local clock, + 2 = monotonic clock, 3 = real clock) processor.max_cstate= [HW,ACPI] Limit processor to maximum C-state diff --git a/include/linux/timekeeping.h b/include/linux/timekeeping.h index ec89d84..0b99a84 100644 --- a/include/linux/timekeeping.h +++ b/include/linux/timekeeping.h @@ -233,6 +233,7 @@ static inline u64 ktime_get_raw_ns(void) extern u64 ktime_get_mono_fast_ns(void); extern u64 ktime_get_raw_fast_ns(void); +extern u64 ktime_get_log_ts(u64 *offset_real); /* * Timespec interfaces utilizing the ktime based ones diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c index 1f4a29b..dfe9ab0 100644 --- a/kernel/printk/printk.c +++ b/kernel/printk/printk.c @@ -420,6 +420,8 @@ static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len, return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len); } +static u64 printk_get_ts(void); + /* insert record into the buffer, discard old ones, update heads */ static int log_store(int facility, int level, enum log_flags flags, u64 ts_nsec, @@ -468,7 +470,7 @@ static int log_store(int facility, int level, if (ts_nsec > 0) msg->ts_nsec = ts_nsec; else - msg->ts_nsec = local_clock(); + msg->ts_nsec = printk_get_ts(); memset(log_dict(msg) + dict_len, 0, pad_len); msg->len = size; @@ -1037,6 +1039,12 @@ static inline void boot_delay_msec(int level) static int printk_time = CONFIG_PRINTK_TIME; +/* + * Real clock & 32-bit systems: Selecting the real clock printk timestamp may + * lead to unlikely situations where a timestamp is wrong because the real time + * offset is read without the protection of a sequence lock in the call to + * ktime_get_log_ts() in printk_get_ts() below. + */ static int printk_time_param_set(const char *val, const struct kernel_param *kp) { @@ -1058,6 +1066,14 @@ static int printk_time_param_set(const char *val, case 'y': printk_time = 1; break; + /* 2 = monotonic clock */ + case '2': + printk_time = 2; + break; + /* 3 = real clock */ + case '3': + printk_time = 3; + break; default: pr_warn("printk: invalid timestamp value\n"); return -EINVAL; @@ -1075,6 +1091,21 @@ static struct kernel_param_ops printk_time_param_ops = { module_param_cb(time, &printk_time_param_ops, &printk_time, S_IRUGO | S_IWUSR); +static u64 printk_get_ts(void) +{ + u64 mono, offset_real; + + if (printk_time <= 1) + return local_clock(); + + mono = ktime_get_log_ts(&offset_real); + + if (printk_time == 2) + return mono; + + return mono + offset_real; +} + static size_t print_time(u64 ts, char *buf) { unsigned long rem_nsec; @@ -1651,7 +1682,7 @@ static bool cont_add(int facility, int level, const char *text, size_t len) cont.facility = facility; cont.level = level; cont.owner = current; - cont.ts_nsec = local_clock(); + cont.ts_nsec = printk_get_ts(); cont.flags = 0; cont.cons = 0; cont.flushed = false; diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c index 34b4ced..698e1a2 100644 --- a/kernel/time/timekeeping.c +++ b/kernel/time/timekeeping.c @@ -43,6 +43,7 @@ static struct { static DEFINE_RAW_SPINLOCK(timekeeper_lock); static struct timekeeper shadow_timekeeper; +static int timekeeping_active; /** * struct tk_fast - NMI safe timekeeper @@ -401,6 +402,16 @@ u64 ktime_get_raw_fast_ns(void) } EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns); +u64 ktime_get_log_ts(u64 *offset_real) +{ + *offset_real = ktime_to_ns(tk_core.timekeeper.offs_real); + + if (timekeeping_active) + return ktime_get_mono_fast_ns(); + else + return local_clock(); +} + /* Suspend-time cycles value for halted fast timekeeper. */ static cycle_t cycles_at_suspend; @@ -1267,6 +1278,8 @@ void __init timekeeping_init(void) write_seqcount_end(&tk_core.seq); raw_spin_unlock_irqrestore(&timekeeper_lock, flags); + + timekeeping_active = 1; } /* time in seconds when suspend began for persistent clock */ diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug index d0caf14..cb2598f 100644 --- a/lib/Kconfig.debug +++ b/lib/Kconfig.debug @@ -1,8 +1,8 @@ menu "printk and dmesg options" config PRINTK_TIME - int "Show timing information on printks (0-1)" - range 0 1 + int "Show timing information on printks (0-3)" + range 0 3 default "0" depends on PRINTK help @@ -13,7 +13,8 @@ config PRINTK_TIME The timestamp is always recorded internally, and exported to /dev/kmsg. This flag just specifies if the timestamp should be included, not that the timestamp is recorded. 0 disables the - timestamp and 1 uses the local clock. + timestamp and 1 uses the local clock, 2 uses the monotonic clock, and + 3 uses real clock. The behavior is also controlled by the kernel command line parameter printk.time=1. See Documentation/kernel-parameters.txt -- 1.8.3.1